Extending autojob¶

autojob is designed to facilitate extension to fit your niche needs. This page outlines what is needed in order to ensure that your extension fits nicely into the autojob ecosystem. In particular, this document covers how to extend the following constructs:

Task
Calculation

Task¶

Writing your own Task requires implementation of

from_directory

If you intend to override Task.from_directory() in your subclass, be careful calling Task.from_directory() with magic_mode = True. Since this parameter may result in an infinite loop. If you are going to support calling from_directory with magic_mode = True considering starting your function with the following if block:

def func():
    builder = TaskMetadata.from_directory(dir_name)._build_class
    if magic_mode:
        return cls.load_magic(dir_name, builder)

This ensures that the loading is deferred to the appropriate subclass and avoids the temptation of passing magic_mode. Note that if you were to construct your override in this way:

...
task = Task.from_directory(dir_name, strict_mode, magic_mode)
# do patching here
...

Then, if _build_class is set in the metadata file, you will effectively call your code twice!

Calculation¶

Useful points for customization:

Calculation.write_python_script: Define a new template file
Calculation.prepare_input_atoms: Define preparation steps in addition to copying magnetic moments or freeze non-adsorbate atoms.
Calculation.get_output_atoms: This function should return an Atoms object Path that represents the directory of the calculation
Calculation.load_calculation_outputs: This function should return a dictionary that maps outputs (string keys) to their values when passed a Path that represents the directory of the calculation
FILES_TO_CARRYOVER: This should be a module-level
constant which defines the files from a previous calculation to carry-over to a new job. Usually, this is the checkpoint-type file employed by the calculation software.